AI Repo | Frontend Repo | Backend Repo

Studio-YAIverse Backend

"Text-guided 3D synthesis by GET3D + NADA"


Car → Police	Car → Sketch

Motorbike → Tiger	Shoe → Mossy

2nd YAI-Con: Studio YAIVerse Team's backend repository

Running Instructions

Getting Dependencies: Python & CUDA

Python 3.8: We highly recommend using python3.8 for compatibility.
CUDA and CUDNN: We need cuda-11.1 and cudnn-8.0.5 for compiling GET3D extensions.
- Manually install via homepage, or use our Dockerfile. (recommended)
- Instead, set TORCH_WITHOUT_CUSTOM_OPS_COMPILE=0 in environ or django settings, to disable GET3D extensions. (little bit slow inference speed)

Runtime Preparation

(Step 1) Recursively clone this repository (due to submodule dependency).

git clone --recursive https://github.com/studio-YAIVERSE/Backend studio-YAIVERSE-Backend
cd studio-YAIVERSE-Backend
# if you are updating an existing checkout
git submodule sync && git submodule update --init --recursive

(Step 2) Make virtual environment. You can use Dockerfile instead. (see below)

python3 -m pip install virtualenv
python3 -m virtualenv venv --python=3.8
source venv/bin/activate

(Step 3) We provide setup.sh for installing dependencies and model weight retrieval.

sh setup.sh

Runtime preparation with Docker

Using Docker image is optional but recommended. Follow this instead of (Step 2) upon.

Build Docker image

docker build -f Dockerfile -t studio-yaiverse:v1 .

Start an interactive docker container

docker run --gpus device=all -it --rm -v $(pwd):/workspace -it studio-yaiverse:v1 bash

How to run dev server

--noreload is required: otherwise, model is loaded twice.
you can use python3 manage.py interface with python3 -m studio_YAIVERSE.

python3 -m studio_YAIVERSE runserver --noreload
TORCH_DEVICE=cuda:0 python3 -m studio_YAIVERSE runserver --noreload  # to specify device
TORCH_ENABLED=0 python3 -m studio_YAIVERSE runserver  # to disable pytorch ops

Appendix. How to deploy with gunicorn & nginx

View / Hide

SERVER_NAME, SECRET_KEY(optional) is required. alternate it to your server address.
```
export SERVER_NAME={your-server-address}
export SECRET_KEY={secret-key}
```

Write secret.json: implement SECRET_KEY, ALLOWED_HOSTS, and DATABASES

echo "{
  \"ALLOWED_HOSTS\": [\"$SERVER_NAME\"],
  \"SECRET_KEY\" : \"$SECRET_KEY\",
  \"DATABASES\": {
    \"default\": {
      \"ENGINE\": \"django.db.backends.sqlite3\",
      \"NAME\": \"$(pwd)/db.sqlite3\"
    }
  }
}
" > secret.json

Write gunicorn service file (gunicorn is already installed by setup.sh)

sudo echo "[Unit]
Description=studio-YAIVERSE gunicorn daemon
After=network.target

[Service]
User=$(whoami)
Group=$(whoami)
WorkingDirectory=$(pwd)
ExecStart=$(which gunicorn) \\
        --workers 2 \\
        --bind unix:/tmp/studio-yaiverse-gunicorn.sock \\
        studio_YAIVERSE.wsgi:application

[Install]
WantedBy=multi-user.target" > /etc/systemd/system/studio-yaiverse-gunicorn.service

Install nginx, prepare static files, and configure your site.

sudo apt install nginx

TORCH_ENABLED=0 python -m studio_YAIVERSE collectstatic

sudo echo "server {
        listen 80;
        server_name $SERVER_NAME;

        location = /favicon.ico { access_log off; log_not_found off; }

        location /static {
                alias $(pwd)/staticfiles;
        }

        location /media {
                alias $(pwd)/attachment;
        }

        location / {
                include proxy_params;
                proxy_pass http://unix:/tmp/studio-yaiverse-gunicorn.sock;
        }
}" > /etc/nginx/sites-available/studio-yaiverse-site

if [ -f /etc/nginx/sites-enabled/default ]; then
    sudo rm /etc/nginx/sites-enabled/default
fi
sudo ln -s /etc/nginx/sites-available/studio-yaiverse-site /etc/nginx/sites-enabled/studio-yaiverse-site

Enable and start gunicorn and nginx service

sudo systemctl enable studio-yaiverse-gunicorn
sudo systemctl start studio-yaiverse-gunicorn
sudo systemctl enable nginx
sudo systemctl restart nginx

Your site is now running at http://$SERVER_NAME!

studio-YAIVERSE / Backend